Streamflow Forecasting on Basins with Limited Hydrologic Data

One essential step in streamflow forecasting for river basins is the choice of suitable values for the parameters of the employed hydrologic model. The classical approach to doing this via model calibration typically requires that considerable amounts of data be collected and assimilated. Therefore, the use of the classic off-line batch optimization methods is limited when hydrologic predictions must be generated for a previously ungaged watershed that has only recently been instrumented thus having only short time series of calibration data available. This presentation introduces the framework for a Bayesian Recursive Estimation (BaRE) approach to hydrologic prediction that is well suited to work with these few measurements.

The BaRE algorithm can be used in an operational setting for simultaneous parameter estimation and one-step-ahead prediction. It computes probabilities of parameter values with as little as one measurement and updates these probabilities recursively, thereby reducing the parameter uncertainty, as more data become available. One-step-ahead predictions are computed simultaneously and are described in terms of the probabilities associated with different output values. The uncertainty in these predictions is the compound effect of the parameter, data and structural uncertainties associated with the underlying model and is gradually improved by the aforementioned reduction of the uncertainty associated with the parameter estimates. The effectiveness and efficiency of the method is illustrated in the context of the complex operational SAC-SMA model, using data from the Leaf River Basin, Mississippi.